Analogy, Expertise, Creativity Flashcards
structure-mapping theory of analogy
analogy: when two conceptual domains share relational similarity, not feature or object-based similarity
e. g. rutherford model of atom
structure-mapping theory:
constraints:
1. structural consistency
- one-to-one mapping
- parallel connectivity
e. g. sun-> nucleus planet ->electron
2. systematicity
- deeply nested relational structures make better analogies (local vs global)
e. g. US invasion of Iraq like WW2
structure-mapping vs feature comparison
One-to-one mapping, parallel connectivity, systematicity
Falkenheiner, Forbus, & Gentner (1989) in Artificial Intelligence describe the computational implementation of this process in ~50 pages
Compare to: S(A,B)= qf(A∩B)−af(A–B)−bf(B–A)
Structure-mapping is computationally expensive
Even comparisons of relational structures of simple scenes perceptual stimuli are difficult when working memory is depleted from e.g., induced anxiety
Representational pluralism
Inference vs. Memory
No one is arguing that all information is only represented in a structured format
Information can be represented differently depending on the process
– e.g, making an analogy to drive inferences with two cases in front of you to compare vs. having a single case in front of you and one must be cued in memory
inference vs memory
gentner et al (1993):
stories of relational and surface match to target story
subjects given all story and asked to evaluate inferences from one story to another based on shared content -> select relational match
subjects given target story after delay and asked which other story does this remind you of -> select surface match
holyoad & koh (1987): greatest reminding of cases that share relational and superficial similarity
model of analogical memory and inference (forbus et al 1995)
cases represented as just an unstructured set of features during memory search
LTM too cavernous for computationally expensive structure mapping
content overlap, represented as vectors cues remindings
hopefully things that have same deep structure will also have same content
inert knowledge problem
the limits of analogical memory
beyond remembering individual analogous cases
classifying problems/phenomena by their underlying relational structure
-abstract relational knowledge helps you to see the structure in novel cases
physics experts vs novices
-categorise problem via underlying principles vs surface properties
test effect Karpicke & Roediger (2006)
4 Conditions of learning foreign language
- Study - Test (ST): every item studied and tested every time
- Study/drop - Test (SnT): once an item is correct on a test, it is dropped from the next study session, but it is still included on every test
- Study – Test/drop (STn): once an item is correct on a test, it is dropped from later test sessions, but kept in study sessions
- Study/drop – Test/drop (SnTn): once an item is correct on a test, it is dropped from both study and test sessions.
result:
- All conditions reached same high accuracy during learning phase 1 week before final test
- after something is recalled accurately, further study doesn’t help, continued testing maintains high performance
- after something is recalled accurately, without further testing, it is forgotten, regardless of further study
shortcoming of test effect
cognitive deficit is un-cued spontaneous analogical transfer or “relational categorisation”
term response vs application questions
transfer quizzes help both learning the definition and further transfer more than not quizzing
definition quizzes only help with learning definition, do not help transfer
test effect summary
• Practicing retrieval with testing is more beneficial than more study (after initial study).
• To get transfer however, transfer must be part of the testing
– Otherwise, just long-term retention. Which isn’t bad, but is not enough.
•Students are unaware of this benefit, potentially because retrieval is difficult/requires effort, and so does not appear to show they have learned it.
Interleaving: Rohrer & Pashler (2010)
• Typical textbooks and curriculum encourage massed/ blocked practice
– For example, many quadratic equation problems in a row
– Think of problems to be solved using the quadratic equation as a
category, and any given problem as an exemplar
– Massing/blocking is dealing with each category one at a time
• However there is lots of evidence now that interleaving benefits learning
– Mixing up examples from many different problem categories
- In massed/blocked practice, you often don’t have to categorize the problem first, you just do it.
- Interleaving forces problem categorization, and solving the problem.
•Most examples of interleaving being beneficial concern categories that are hard to discriminate from each other
– “the integration problems: ∫exedx and ∫xexdx resemble each other yet require different techniques (the latter requires integration by parts).”
•Transfer is often about problem categorization
– Knowing what kind of problem you are facing tells you what solution procedure to use
– Students often “know” solution procedures without knowing when to apply them
exception of desirable difficulty
Work from Rob Goldstone’s lab has shown that sometimes students self-selecting a blocking strategy can be beneficial, especially when within-category items are rather distinct from each other.
John Sweller has argued that desirable difficulties are not desirable when the material is so complex that it intrinsically maxes out working-memory capacity. Then any extra difficulty from learning-mode is harmful
The relational shift in child development and expertise development
• Relational concepts often have correlated superficial features
– e.g., predators typically have big teeth
– textbooks often correlate physical/mathematical principles with story content in word problems (Mayer, 1982)
• Learners shift their focus from superficial features to focus on more abstract relational commonalities with increased experience
– e.g., uncle means “bearded guy who brings presents” before it means sibling of a parent (Keil, 1984)
– Novice physics students categorize problems as “problems with pulleys” while expert students categorize as “problems about the conservation of momentum” (Chi, Feltovich, & Glaser, 1981)
• In both children and adult category learning tasks, comparison helps shift towards relational structure
Relational shift in undergraduate science: causal system categories
Examining novice-expert shift
Do college students recognize and use causal system categories?
Novices: Social science majors
Experts: Physical science majors
Given cards to sort into different piles with descriptions of phenomena on them that vary in their domain and causal system
Social science students sort by domain
Physical science students sort by causal system
biggest challenge in education? conceptual change
in science education, we don’t just learn new scientific relational concepts, but we have to transform our pre- conceptions about how the world works to the “scientifically correct” model
Gadgil et al. argue that mental models are about the system of relations among the elements in a domain
– Like “Theory Theory” says
learning the correct circulatory system: expert explain vs compare and contrast
compare and contrast group performs better